Panel PPML structural gravity estimation - raw vs Absolute distance

[Kostas Dellis]

Dear Stata users

I am using the ppml_panel_sg command (Zylkin et al., 2018) to measure the effect of institutional distance on bilateral FDI flows using bilateral FDI flows across 36 economies 2005-2016.

* Example generated by -dataex-. For more info, type help dataex
clear
input str3(host_countrycode or_countrycode) float year str6 pair float(fdi_res ldist) byte(comlang smctry) float comrelig byte fta_wto float(inst_dif inst_dif_ab)
"AUS" "GRC" 2009 "AUSGRC" . 9.638067 0 0 .001449 0 -.08333325 .08333325
"AUS" "SVK" 2009 "AUSSVK" . 9.675989 0 0  .23878 0 -.11111093 .11111093
"AUS" "ITA" 2009 "AUSITA" . 9.700958 0 0 .247214 0        .75       .75
"AUS" "CHE" 2009 "AUSCHE" . 9.721559 0 0 .257814 0   .8333335  .8333335
"AUS" "SVN" 2009 "AUSSVN" . 9.689849 0 0 .211374 0   .3888891  .3888891
"AUS" "KOR" 2009 "AUSKOR" . 9.027577 0 0 .040214 0   .7777779  .7777779
"AUS" "ISL" 2009 "AUSISL" . 9.719113 0 0 .229082 0 -1.9166665 1.9166665
"AUS" "NLD" 2009 "AUSNLD" . 9.720653 0 0 .225756 0  1.0277779 1.0277779
"AUS" "USA" 2009 "AUSUSA" . 9.680893 1 0 .191276 1  -.1944444  .1944444
"AUS" "CHL" 2009 "AUSCHL" .   9.3373 0 0 .247481 1         -2         2
"AUS" "TUR" 2009 "AUSTUR" . 9.613156 0 0  .00228 0  -.8333333  .8333333
"AUS" "IRL" 2009 "AUSIRL" . 9.754436 1 0 .284673 0  .13888907 .13888907
"AUS" "LUX" 2009 "AUSLUX" . 9.722378 0 0   .2781 0   .3888891  .3888891
"AUS" "ISR" 2009 "AUSISR" .   9.5604 1 0  .00359 0  -1.222222  1.222222
"AUS" "HUN" 2009 "AUSHUN" . 9.667499 0 0 .210304 0  1.1111112 1.1111112
end

I am using time-varying host and origin fixed effects for parsimony.

 global grav1 ldist comlang smctry comrelig
ppml_panel_sg fdi_res `v' $grav1  fta_wto eps_dif_ab, ex(or) im(host) y(year) nopair cluster(pair2)

While there is no issue with the Absolute Institutional Difference inst_dif_ab, when the raw difference inst_dif (including negative values) is included I get the following message

eps_dif appears to be collinear with your set of fixed effects

Is there a way to fix this? The code works once I include time-varying host and origin variables in the RHS

KD